Analyzing WashingtonState'sWelfare Program Design, WorkFirst
Dan Axelsen
ERS Group
850-562-1211
Hal W. Snarr
Department of Economics and
Transportation/Logistics
School of Business and Economics
North CarolinaState A & T
336-334-7744
Abstract
Much debate in the early nineties centered on whether the federal entitlement program Aid for Families with Dependent Children (AFDC) reduced welfare dependency. Many contend that AFDC discouraged work, increased welfare dependency, and undermined the institution of family. Partly in response to these criticisms, welfare was reformed through the Personal Responsibility and Work Reconciliation Act (PRWORA) in 1996. PRWORA modified the primary objectives of welfare by placing more emphasis on work experience accumulation and less on human capital accumulation. Temporary Assistance for Needy Families (TANF) was designed to meet this primary objective. WashingtonState’s TANF program, WorkFirst, utilizes a progressive system of programs (components) aimed at reducing welfare dependency through labor force participation. WorkFirst components have a variety of objectives including skills training, temporary subsidized employment, and mentoring. WorkFirst’s objective is to accumulate work experience of welfare recipients, thus making them more employable. More work experience should place upward pressure on wage rates, which then in turn reduces welfare dependency. We analyze the working decision as it is related to WashingtonState’s program design using a binary choice probit model. We find that welfare recipients who are enrolled in the later stage components of WorkFirst are more likely to find work and exit welfare than those that have only completed the initial components designed under WorkFirst. Cumulatively, WorkFirst seems to be an effective welfare program design.
Introduction[1]
On July 1, 1997 the Temporary Assistance for Needy Families (TANF) program replaced the federal entitlement program, Aid for Families with Dependent Children (AFDC) in WashingtonState. TANF provides temporary cash assistance and job training to low-income families with dependent children. Each State is allowed to tailor their welfare program design to the specific needs of their clients. In August 1997, the State of Washington created WorkFirst as its welfare program designed under TANF. The WorkFirst design discourages welfare dependency through the promotion of labor force participation and sanctions for non-compliance. Programs (components) under WorkFirst provide individuals with mentors, educational opportunities, and paid work experience. The underlying premise of WorkFirst is that work experience accumulation places upward pressure on wage rates. Components within WorkFirst promote work experience and skill building, and through the completion of cumulative stages within WorkFirst welfare recipients become less welfare dependent, and in many instances self-sufficient.
WashingtonState’s WorkFirst program is not an entitlement; it provides temporary cash assistance, job training, and work experience to welfare recipients. Recipients are allowed a maximum of 60 months of lifetime cash assistance, with at most 24 consecutive months of assistance. Recipients must work or search for employment to receive cash assistance.[2] Unsuccessful individuals not finding employment but making an effort according to WorkFirst guidelines can qualify for extensions. Because physically or mentally disabled individuals cannot participate in WorkFirst, they will continue to receive welfare assistance.[3]
WorkFirst is a comprehensive and progressive welfare-to-work program that utilizes many components consisting of over 30 different employment induced welfare programs. For the purposes of this paper, we will study the effects of only five of these WorkFirst components. They include Job Success Coach Initiative (JSCI), WorkFirst Post-employment Labor Exchange (WPLEX), Job Components (JC), Pre-Employment Training (PET), and Community Jobs (CJ). We chose these components because they are employment training, counseling and guidance programs. All five components are different in their nature and each attracts a different clientele. They are also progressive and attract the most political attention.[4]
Clients with very poor job skills often enter the WorkFirst program by enrolling in CJ, PET or JC. The CJ program is a community-based work and skill-building experience for heads of households receiving TANF benefits. The CJ program provides participants with valuable work experience and skills training. CJ benefits both the individual and local communities, often leading to a permanent job and job retention while meeting WorkFirst participation requirements “to work, look for work or prepare for work.” Recipients that secure jobs under CJ are only temporarily employed, most typically for six months. PET is a program designed to assist individuals with additional education. The majority of PET individuals are enrolled in a Microsoft funded community college program focused on the acquisition of computer skills. The JC program is a series of classes offered by the state. These classes teach participants how to act professional, look for work, and interview for jobs.
The JSCI and WPLEX programs tend to attract stronger clients whom are typically in the final stages of the WorkFirst program. The JSCI program assigns a job coach to each enrollee. The job coach guides individuals in their job search journey by offering advice and counseling. After clients have obtained employment they receive additional counseling from WPLEX call center associates. After enrollees enter the work force, the WPLEX program can assist them in finding better jobs as their job-skills improve and work experience increases.
WorkFirst is a progressive welfare-to-work design because clients typically start in JC, CJ or PET. After completion of any of these preliminary components, clients enroll in JSCI or WPLEX.
Literature Review
Much experimental design was conducted in the early 1990's as an application of welfare reform. Waiver programs allowed states to experiment with alternatives to the traditional AFDC design.[5] Many of these waiver programs resemble TANF, focusing on strong work requirements through job search, therefore the waiver period can be thought of as a foundation for understanding how to model TANF correctly. The waiver period and the research associated with this period directly evaluate different program designs, unified by the imposition of strong work requirements. Friedlander and Hamilton (1996) examined an experimental welfare model that would eventually mimic TANF by testing whether imposing an obligation to work, and the implementation of these policies that lead to work, actually increase employment levels which in turn should reduce welfare dependency. The experimental data and design used by Friedlander and Hamilton (1996) is known as the Saturation Work Initiative Model (SWIM). From 1985 to 1987 the county of San Diego initiated the SWIM welfare design, which took a cohort of people currently enrolled in AFDC and placed them in work assignments and other types of employment-directed programs. Results show 75 percent of those who participated in SWIM were employed compared to 68 percent for the control group, a difference statistically significant at the 5 percent level. Friedlander and Hamilton (1996) successfully find that the SWIM model reduces welfare dependency more than the existing AFDC design. However, the authors' do not conclude that the waiver program is a successful overall welfare design, it is only marginally better than AFDC.
Bloom and Michalopoulos (2001), Fein et al. (2001) and O’Neill and Hill (2001) all look at the effect that similar waiver programs have had on employment, program participation, and wages. Each of these papers found waivers to have positive effects on employment and wages. In contrast, O’Neill and Hill, Bloom and Michalopoulos, and Fein et al. are all similar in that they find a negative effect on AFDC participation. They suggest that the effects on wages and employment are due to the enforcement of work requirements, not from marginal incentive changes. Similar results are found by Bartik and Eberts (1999); Blank (2000); Figlio and Ziliak (1999); Wallace and Blank (1999); and Ziliak, Figlio, Davis, and Connelly (2000). Each of these authors' fails to determine whether each program design works as it is intended.
One issue with studying waiver programs is that only short-term effects of program reform are considered. Friedlander, Greenberg, and Robins (1997) evaluate the long-term aspects of government programs for the disadvantaged. Since welfare’s initial inception in 1935, they find no support for positive long-term effects on employment, wages, poverty, income inequality, and welfare participation. Although most evaluations of the welfare population conclude that there are positive, significant effects of training programs in the short-run, they find that there is no evidence to support this claim over the longer-term. One explanation is the skills learned by participants are not valued by employers over the long-term. They should evaluate such a claim given the overall shape of the particular labor market for which their sample resides.
Supply controls typically include recipients’ marital status, age, race, education, recent work experience, and welfare history. Surveys are the most common method for collecting labor supply data. The most commonly used surveys include the Survey of Income and Program Participation (SIPP), the Panel Study on Income Dynamics (PSID), the National Longitudinal Survey of Youth (NLSY) and Longitudinal Database of Cases (LDB).[6]
Hoynes (2000) incorporated panel data from a subset of the LBD survey to estimate the probability of exits, returns and duration with respect to labor supply and demand variables using a discrete time hazard model. She found that labor market conditions significantly determined exits, returns and duration between 1987 and 1992. That is, welfare recipients are more likely to exit welfare and less likely to return when they are earning more. Additionally, she found that the market wage was a strong determinate of welfare dependency.
Harris (1993) uses a panel of data, from the PSID survey, to evaluate single mothers enrolled on AFDC from 1984 to 1986. She finds that the guarantee level is insignificant, and that a mother enrolled on AFDC exits approximately 67 percent of the time because of an employment opportunity.[7] Harris (1996) follows up her own work by looking at a longer panel from the same survey to estimate the probability of single mothers re-entering AFDC once they have exited. Once again, Harris finds the guarantee to be insignificant and, in general, claims that analyzing welfare is not always about the recipient weighing the costs and benefits of exiting and then returning. More important factors are those such as age, number of children, education and marriage. These factors significantly influence repeat dependency. The choice to exit and not return is mostly a function of the single mother’s need for dependency and the existing social context she lives in. Women receiving welfare assistance are less likely to exit if they are single, have more dependents, older and less educated.
In a preliminary study using data from the Washington State Population Survey, Learch, Mayfield, and Burley (1999) found that Washington State females who participated in WorkFirst job search programs had 12 to 27 percent higher employment rates, earned approximately $213 more per month, and had between 14 and 20 percent less welfare enrollment than clients not participating. They used client characteristics as labor supply controls. Labor demand controls included local economic variables and local welfare office administrative practices. Their results were admittedly preliminary and required verification. We hope to verify these results using administrative records, and more importantly, we intend to analyze the WorkFirst design in the manner for which it is designed.
Supply side factors tend to be significant predictors of welfare dependency participation, and so they should be included in any empirical specification. However, generally speaking, demand side factors only generate statistically significant coefficients in studies using panels over long periods of time. Because our time horizon is so short, we do not include demand side factors in this analysis that are aggregated at the state or local level.
The goals of the above studies are only partially relevant to our work. However, our methodology is consistent with past methodologies. Typical methodologies analyze the work-decision with a binary dependent variable (probit and logit models). The use of a dichotomous variable captures whether or not recipients are working. Any model using this feature has the ability to predict labor force participation on a vector of independent variables. These types of models are also capable of evaluating welfare dependency, indicated by welfare exits and reentry.
Many current welfare designs have a progressive, cumulative nature to them. By modeling progression one can partially determine how well a program is designed in terms of effectiveness in reducing welfare dependency. To do this, one must be able to collect supply side data specific to individuals receiving welfare benefits and their component participation choices. To categorize the sample in this manor, one must have access to administrative records. Individuals entering WorkFirst progress through different components of the program until they are ready to exit and enter the work force. Participants that do not complete the series of components should be more dependent on welfare and less likely to enter the labor force. The literature has not yet captured this progressive analysis.
Data Set
Most welfare studies use survey data. We use Washington State Administrative records for our analysis. We have merged wage data from Unemployment Insurance (UI) wage files with WorkFirst welfare data from the JOBS Automation System (JAS) of the Employment Security Department (ESD).[8] The merged data produces a panel data set, consisting of both cross-sectional and time-series elements specific to individuals enrolled in WorkFirst. Furthermore, this panel provides us the ability to capture both the working decision and welfare enrollment decision of each individual recipient over time.[9] We end up with a cross-section of welfare recipients between 2000 and 2001.
The WorkFirst design can only be analyzed by using administrative records. Surveys offer information about the aggregate welfare population and are often inaccurate due to survey approaches. By using individual welfare and wage data we can track individual decision making over time. Specifically, we can track which WorkFirst components individuals have enrolled in, which component they started in, when they finished each component, and how they are doing in the labor force after exiting each component. The merged data we have been able to collect allows us to evaluate what no other study has been able to, namely, measure the effectiveness of the WorkFirst design.
Table 1 provides various descriptive statistics of explanatory factors for the entire sample. Of the five WorkFirst components, WPLEX generated the highest median earnings and employment rates while having the lowest percentage of reentry into WorkFirst; $2,742 per quarter, 92.1 percent, and 12.9 percent respectively. JSCI has the highest exit percentage at 23 percent. JC, PET, and CJ appear to be the least successful of the WorkFirst components. For example, after completing CJ only 45.3 percent were employed, only 12 percent exited, and median earnings were a meager $1,624 per quarter. Demographically, the components appear very similar; marriage percentages, education levels, mean age of youngest child, and number of children residing in the welfare recipients household are similar across components. However, work experience is very different across components. Individuals enrolled in WPLEX and JSCI have more work experience than those in JC, PET, and CJ. Even though the performance of these later programs appears to be disappointing, they are not. Through self-selection, participants that enter JSCI and WPLEX are more prepared for work. Individuals that are less prepared for the work force are encouraged to first enter JC, PET, and CJ.
We speculate that upon completion of these initial components, individuals are more likely to succeed in WorkFirst’s terminal components, WPLEX and JSCI. It is evident from Table 1 that enrollees who successfully progress through WorkFirst components are better suited for the work force. That is, the likelihood an enrollee finds a job and exits welfare is higher than that of an individual who bypasses one or more of the initial components of WorkFirst. Those enrolled in the later stage components seem to be descriptively more successful than those enrolled in the initial stages of the WorkFirst design.
TABLE 1 Descriptive StatisticsComponent / JSCI / WPLEX / JC / PET / CJ
N / 691 / 1122 / 8637 / 1588 / 1838
N with UI-reported wages / 571 / 1033 / 4635 / 739 / 832
% employed in the CY2001, Q1 / 83.00% / 92.10% / 53.70% / 46.50% / 45.30%
Median quarterly earnings in CY2001, Q1 / $1,657 / $2,724 / $1,496 / $1,954 / $1,624
% who exit TANF in January 2001 / 23.00% / 17.90% / 18.00% / 14.70% / 12.00%
% returning in Feb., March, or April 2001 / 15.10% / 12.90% / 17.10% / 19.30% / 18.20%
Mean age of client / 31* / 33 / 33 / 32* / 33*
Mean age of youngest child / 6* / 7* / 8* / 7* / 8*
Mean number of children / 2 / 2 / 2 / 2 / 2
% that never made it through eighth grade / 10.48% / 11.42% / 12.34% / 10.45% / 12.01%
% with some high school / 41.22% / 24.28% / 29.98% / 34.54% / 35.20%
% with high school diploma/GED / 42.16% / 49.31% / 45.68% / 44.37% / 43.23%
% with some college / 5.50% / 13.17% / 10.36% / 9.64% / 8.70%
% with at least a BS/BA / 0.62% / 1.81% / 1.64% / 0.99% / 0.85%
Percent married / 23% / 32%* / 28%* / 24%* / 22%*
% of past 24 quarters employed / 43.79% / 44.13% / 34.90% / 33.05% / 35.83%
% of past 4 quarters employed / 57.58% / 62.10% / 43.02% / 37.48% / 30.16%
Median earnings over past 4 quarters / 1,018 / 1,680 / 878 / 895 / 681
Percent with zero earnings in 6 years / 2.28% / 4.18% / 12.79% / 11.83% / 10.64%
Data Source: JAS, UI wage file, monthly JSCI report, an * says that the cell in question is significantly different than the caseload.